Part of the retreat Hackathon.
---------
Co-authored-by: David Thrane Christiansen <david@davidchristiansen.dk>
Co-authored-by: Mario Carneiro <di.gama@gmail.com>
The Canonicalizer creates a “key” expression eliding certain information
(implicit parameters, levels), and `getFunInfo` can be
confused by these terms (in particular, wrong number of level
parameters).
By running `getFunInfo` on the original expression we avoid this, and
can just put `[]` as the level list in the key.
This changes how Nat typeclass checks in offset terms from syntactic
equality to definitional equality with "instances" transparency.
This may have a negative performance penalty in `isOffset?`, but it
should be small in common cases since the relevant instances are small
terms.
This closes#3836
This removes simp attributes from `Nat.succ.injEq` and
`Nat.succ_sub_succ_eq_sub` to replace them with simprocs. This is
because any reductions involving `Nat.succ` has a high risk of leading
proof performance problems when dealing with even moderately large
numbers.
Here are a couple examples that will both report a maximum recursion
depth error currently. These examples are fixed by this PR.
```
example : (123456: Nat) = 12345667 := by
simp
example (x : Nat) (p : x = 0) : 1000 - (x + 1000) = 0 := by
simp
```
The matches returned by the lazy discriminator tree are partially
constrained by a priority, but ties are broken by the order in which
keys are traversed and the order of declarations.
This PR changes the match key traversal to use an explicit stack rather
than recursion and implicitly changes the order in which results are
returned to favor left-matches first e.g., given the term `f a b` with
constants `f a b`, and a tree with patterns `f a x -> 1` `f x b -> 2`
that have the same priority, this will return `#[1, 2]` since the early
matches for the key `a` are returned before the match for `x` which has
a star.
This appears to address the [lower quality results mentioned on
zulip](https://leanprover.zulipchat.com/#narrow/stream/428973-nightly-testing/topic/Mathlib.20status.20updates/near/429955747).
When `discharge?` failed, the `usedSimps` was being restored, but the
cache wasn't. This bug was exposed by issue #3710.
This PR makes the following changes:
- We restore the `cache` at `discharge?`. We use `SMap` to ensure the
operation is efficient.
- We don't need the field `dischargeDepth` anymore at `Simp.Result`.
- `UsedSimps` should use `PHashMap` since it is not used linearly.
closes#3710
---------
Co-authored-by: Mario Carneiro <di.gama@gmail.com>
Now, only `(<- ...)`s occurring in the condition of a pure if-then-else
are lifted.
That is, `if (<- foo) then ... else ...` is ok, but `if ... then (<-
foo) else ...` is not. See #3713closes#3713
This PR also adjusts this repo. Note that some of the `(<- ...)` were
harmless since they were just accessing some
read-only state.
It have to keep it as a private definition for now. We currently only
support duplicate theorems in different modules. Splitters are generated
on demand, and are also used to write code.
This PR includes the following fixes:
- Reserved name resolution inside namespaces
- Equation theorems for `match`er declarations are not private anymore
- Equation theorems for `match`er declarations are realizable
- `foo.match_<idx>.splitter` is now a reserved name
When it was upstreamed, it lost the mention of "revert/intro pattern",
which is helpful for finding this function. Also extended the
description of the function and clarified some points.
---------
Co-authored-by: Scott Morrison <scott.morrison@gmail.com>
Previously:
If the `rfl` macro was going to fail, it would:
1. expand to `eq_refl`, which is implemented by
`Lean.Elab.Tactic.evalRefl`, and call `Lean.MVarId.refl` which would:
* either try kernel defeq (if in `.default` or `.all` transparency mode)
* otherwise try `IsDefEq`
* then fail.
2. Next expand to the `apply_rfl` tactic, which is implemented by
`Lean.Elab.Tactic.Rfl.evalApplyRfl`, and call `Lean.MVarId.applyRefl`
which would look for lemmas labelled `@[refl]`, and unfortunately in
Mathlib find `Eq.refl`, so try applying that (resulting in another
`IsDefEq`)
3. Because of an accidental duplication, if `Lean.Elab.Tactic.Rfl` was
imported, it would *again* expand to `apply_rfl`.
Now:
1. Same behaviour in `eq_refl`.
2. The `@[refl]` attribute will reject `Eq.refl`, and `MVarId.applyRefl`
will fail when applied to equality goals.
3. The duplication has been removed.
[Before](https://github.com/leanprover/lean4/files/14772220/oi.pdf) and
[after](https://github.com/leanprover/lean4/files/14772226/oi2.pdf).
This gets `ByteArray`, `String.Extra`, `ToString.Macro` and `RCases` out
of the imports of `omega`. I'd hoped to get `Array.Subarray` too, but
it's tangled up in the list literal syntax. Further progress could come
from make `split` use available `Decidable` instances, so we could pull
out `Classical` (and possibly some of `PropLemmas`).
I think this was in error in my original Mathlib implementation. We're
not interested in relations other than `=`, so there is no point uses
`MVarId.applyRfl`, which just looks up `@[refl]` tagged lemmas and tries
those.
In a separate PR, I will change `MVarId.applyRfl` so it has a flag to
control whether on `=` it should just hand-off to `MVarId.refl`, or
fail. Failure is appropriate in the version we call from the `rfl`
macro, to avoid doing a double `IsDefEq` check on every `rfl`!
This makes several changes to rw? and lazy discrimination trees based on
test failures in rewrite search.
Changes include:
1. Reverting to Mathlib function for candidate lemma priority in rw?
2. Introducing additional filters for auto-generated named in lazy
discriminator tree.
3. Refactoring lazy discriminator values to clarify what is stored.
4. Including star keys in calculation of match closeness in
prioritization.
5. Using more fields in current core context when initializing lazy
discriminator tree and avoiding max heartbeat issues.
---------
Co-authored-by: Scott Morrison <scott.morrison@gmail.com>
no need to enter `derive_functional_induction` anymore.
(Will remove the support for `derive_functional_induction` after the
next stage0 update, since we are already using it in Init.)
This extends `derive_functional_induction` to work with structural
recursion as well.
It produces the less general, more concrete induction rule where the
induction hypothesis is
specialized for every argument of the recursive call, not just the the
one that the function
is recursing on.
Care is taken so that the induction principle and it's motive take the
arguments in the same
order as the original function.
While I was it, also makes sure that the order of the cases in the
induction principle matches
the order of recursive calls in the function better.
---------
Co-authored-by: David Thrane Christiansen <david@davidchristiansen.dk>
Co-authored-by: Leonardo de Moura <leomoura@amazon.com>
Given a definition `foo`, they were previously called `foo._unfold`
until 4.7.0. We tried to rename them to `foo.def`, but it created too
many issues in the Mathlib repo. We decided to rename it again to
`foo.eq_def`. The new name is also consistent with the `eq_<idx>`
theorems generated for different "cases". That is, `foo.eq_def` is the
equality theorem for the whole definition, and `foo.eq_<idx>` is the
equality theorem for case `<idx>`.
cc @semorrison
This updates the rw? tactic from Mathlib to use lazy discriminator trees
and upstreams it.
---------
Co-authored-by: Scott Morrison <scott.morrison@gmail.com>
This coercion caused difficult-to-diagnose bugs sometimes. Because there
are some situations where converting a string to a name should be done
by parsing the string, and others where it should not, an explicit
choice seems better here.
---------
Co-authored-by: Mac Malone <tydeu@hatpress.net>
This is a rewrite of the `UnusedVariables` lint to inline and simplify
many of the dependent functions to try to improve the performance of
this lint, which quite often shows up in perf reports.
* The mvar assignment scanning is one of the most expensive parts of the
process, so we do two things to improve this:
* Lazily perform the scan only if we need it
* Use an object-pointer hashmap to ensure that we don't have quadratic
behavior when there are many mvar assignments with slight differences.
* The dependency on `Lean.Server` is removed, meaning we don't need to
do the LSP conversion stuff anymore. The main logic of reference finding
is inlined.
* We take `fvarAliases` into account, and union together fvars which are
aliases of a base fvar. (It would be great if we had `UnionFind` here.)
More docs will be added once we confirm an actual perf improvement.
---------
Co-authored-by: Sebastian Ullrich <sebasti@nullri.ch>
These are used in Mathlib's `congr!` and `convert` tactics, which will
be upstreamed soon.
---------
Co-authored-by: Kyle Miller <kmill31415@gmail.com>
using the `substVars` tactic on the goal can remove too much
information, as it does not take into account that the `motive` may
depend on the fixed parameters.
This is fixed by etracting `substVar` from `subst` which expects the
`x`, not the `h : x = rhs`, and then using this tactic on the local
declarations _after_ the `motive` exclusively.
- Add support for reserved declaration names. We use them for theorems
generated on demand.
- Equation theorems are not private declarations anymore.
- Generate equation theorems on demand when resolving symbols.
- Prevent users from creating declarations using reserved names. Users
can bypass it using meta-programming.
See next test for examples.
Before, the termination argument as inferred by `GuessLex` was passed
further
on as `Syntax`, to be elaborated later in `WF.Rel`.
This didn’t feel quite right anymore. In particular if we want to teach
`GuessLex` about guessing more complex termination arguments like
`xs.size -
i`, using `Expr` here is more natural.
So this introduces `TerminationArgument` based on an `Expr` to be used
here.
A side-effect of how the termination arguments are elaborated is that
the unused
variables linter will now look at `termination_by` variables, and that
parameters
past the colon are not even invisibly in scope, so `‹_›` will not find
them
See https://github.com/leanprover-community/mathlib4/pull/11370/files
for examples
of fixing these changes.
This replaces a few uses of initialize with builtin_initialize, and
removes some unneeded functionality added when it was unclear if lazy
discriminator trees would be efficient enough.
This introduces the `ArgsPacker` module and abstraction, to replace the
exising `PackDomain`/`PackMutual` code. The motivation was that we now
have more uses besides `Fix.lean` (`GuessLex` and `FunInd`), and the
code was spread in various places.
The goals are
* consistent function naming withing the the `PSigma` handling, the
`PSum` handling, and the combined interface
* avoid taking a type apart just based on the `PSigma`/`PSum` nesting,
to be robust in case the user happens to be using `PSigma`/`PSum`
somewhere. Therefore, always pass an `arity` or `numFuncs` or `varNames`
around.
* keep all the `PSigma`/`PSum` encoding logic contained within one
module (`ArgsPacker`), and keep that module independent of its users (so
no `EqnInfos` visible here).
* pick good variable names when matching on a packed argument
* the unary function now is either called `fun1._unary` or
`fun1._mutual`, never `fun1._unary._mutual`.
This file has less heavy dependencies than `PackMutual` had, so build
parallelism is improved as well.
Replaces `@[eliminator]` with two attributes `@[induction_eliminator]`
and `@[cases_eliminator]` for defining custom eliminators for the
`induction` and `cases` tactics, respectively.
Adds `Nat.recAux` and `Nat.casesAuxOn`, which are eliminators that are
defeq to `Nat.rec` and `Nat.casesOn`, but these use `0` and `n + 1`
rather than `Nat.zero` and `Nat.succ n`.
For example, using `induction` to prove that the factorial function is
positive now has the following goal states (thanks also to #3616 for the
goal state after unfolding).
```lean
example : 0 < fact x := by
induction x with
| zero => decide
| succ x ih =>
/-
x : Nat
ih : 0 < fact x
⊢ 0 < fact (x + 1)
-/
unfold fact
/-
...
⊢ 0 < (x + 1) * fact x
-/
simpa using ih
```
Thanks to @adamtopaz for initial work on splitting the `@[eliminator]`
attribute.
Remark: this commit removes the `jason1.lean` test. Motivation: It
breaks all the time due to changes we make, and it is not clear anymore
what it is testing.
---------
Co-authored-by: Joachim Breitner <mail@joachim-breitner.de>
this makes the ugly `fst`/`snd` variable names in the functional
induction principles go away.
Ironically I thought in order to fix these name, I should touch the
mutual/n-ary argument packing code used for well-founded recursion, and
embarked on a big refactor/rewrite of that code, only to find that at
least this particular instance of the issue was somewhere else. Hence
breaking this into its own PR; the refactoring will follow (and will
also improve some other variable names.)
